Expressive facial subspace construction from key face selection

MoCap-based facial expression synthesis techniques have been applied to provide CG character with expressive and accurate facial expressions [Deng et al. 2006: Lau et al. 2007]. The representative performance of these techniques depends on the variety of captured facial expressions. It is also difficult to guess what expressions are needed to synthesize expressive face before capture. Therefore, much MoCap data are required to construct a subspace employing dimensional compression techniques, and then the space enables us to synthesize expressions with linear-combination of basis vectors of the space. However, it is hard work to take much facial MoCap data to obtain expressive result.